[BUG]: CUB 2.4.0 fails with multiple definition errors in .cpp files

[valgur]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Compile-time Error

Component

CUB

Describe the bug

Including cub/util_debug.cuh (or any others that use it) from multiple .cpp files causes linking to fail with:

/usr/bin/ld: b.o: in function `cub::CUB_200400___CUDA_ARCH_LIST___NS::Debug(cudaError, char const*, int)':
b.cpp:(.text+0x0): multiple definition of `cub::CUB_200400___CUDA_ARCH_LIST___NS::Debug(cudaError, char const*, int)'; a.o:a.cpp:(.text+0x0): first defined here

This only happens with CUB 2.4.0 from CUDA 12.5.40. CUB 2.3.2 from CUDA 12.4 and older versions work fine.

I am using GCC 11 as the compiler and linker.

How to Reproduce

1. Create files a.cpp and b.cpp with #include <cub/util_debug.cuh> as their contents.
2. Run

g++-11 -I/usr/local/cuda-12.5/targets/x86_64-linux/include -c a.cpp -o a.o
g++-11 -I/usr/local/cuda-12.5/targets/x86_64-linux/include -c b.cpp -o b.o
g++-11 a.o b.o /usr/local/cuda-12.5/targets/x86_64-linux/lib/libcudart.so -o libx.a

Fails with

/usr/bin/ld: b.o: in function `cub::CUB_200400___CUDA_ARCH_LIST___NS::Debug(cudaError, char const*, int)':
b.cpp:(.text+0x0): multiple definition of `cub::CUB_200400___CUDA_ARCH_LIST___NS::Debug(cudaError, char const*, int)'; a.o:a.cpp:(.text+0x0): first defined here
/usr/bin/ld: /usr/lib/gcc/x86_64-linux-gnu/11/../../../x86_64-linux-gnu/Scrt1.o: in function `_start':
(.text+0x24): undefined reference to `main'
collect2: error: ld returned 1 exit status

Expected behavior

No linker errors.

Reproduction link

No response

Operating System

Ubuntu Linux

nvidia-smi output

No response

NVCC version

No response

[valgur]

For context, I encountered this when trying to build these source files:

• https://github.com/puzzlepaint/camera_calibration/blob/00ef4deca274377f00c3654b75b857dd20382cc2/applications/camera_calibration/src/camera_calibration/feature_detection/cuda_refinement_by_matching.h#L32
• https://github.com/puzzlepaint/camera_calibration/blob/00ef4deca274377f00c3654b75b857dd20382cc2/applications/camera_calibration/src/camera_calibration/feature_detection/cuda_refinement_by_symmetry.h#L32

[alliepiper]

I believe that has been fixed in main by PR #1643.

The relevant change is the addition of inline to this line

Can you try making this change locally and see if it resolves it for you?

[valgur]

Setting _CCCL_FORCEINLINE to inline fixed the issue. Thank you!

Glad to see that the fix has already been applied as well.